Extending the representation of classifier conditions part I: from binary to messy coding

We present the first part of a study on the alternative representations of classifier conditions. We introduce an extension of the XCS classifier system in which variable-length messy chromosomes replace the original bitstring representation. With a series of experiments we show that to reach optimal performance covering, matching, and mutation must be adequately defined in order to avoid overgeneralization due to underspecification of classifier conditions.